Generate M&A due diligence PDF reports with LlamaIndex, OpenAI, Pinecone, and S3

Due diligence at machine speed. This n8n workflow reads financial documents with LlamaIndex, embeds them into Pinecone, and produces comprehensive AI-driven reports with GPT-5-mini. Manual review that took days now takes minutes, with consistent, data-backed output.

Last updated: September 2026.

Why Use This Workflow?

Time Savings: Document review and report generation drop from days to minutes.
Cost Reduction: Far less dependence on expensive human analysts for initial extraction and summaries.
Error Prevention: AI-driven extraction stays consistent, reducing human error and oversight.
Scalability: Multiple documents and deals processed in parallel, scaling with your business.

Ideal For

• Investment Analysts & Private Equity Firms: Evaluate targets fast by automating extraction of financials, risks, and business models from deal documents.
• M&A Advisors: Preliminary due diligence with comprehensive overviews, minus the manual effort.
• Financial Professionals: Accelerate research on filings, presentations, and market reports for critical decisions.

How It Works

1. Trigger: A webhook receives multiple due diligence documents (PDFs, DOCX, XLSX) along with associated metadata.
2. Document Processing & Cache Check: Files are split individually. The workflow first checks Pinecone to see if the deal's documents have been processed before (cache hit). If so, it skips parsing and embedding.
3. Data Extraction (LlamaIndex): For new deals, each document is sent to LlamaIndex for advanced parsing, extracting structured text content.
4. Vectorization & Storage: The parsed text is then converted into numerical vector embeddings using OpenAI and stored in Pinecone, our vector database, with relevant metadata.
5. AI-Powered Analysis (Langchain Agent): An n8n Langchain Agent, acting as a "Senior Investment Analyst," leverages GPT-5-mini to query Pinecone multiple times for specific information (e.g., company profile, financials, risks, business model). It synthesizes these findings into a structured JSON output.
6. Report Generation: The structured AI output is transformed into an HTML report, then converted into a professional PDF document.
7. Secure Storage & Delivery: The final PDF due diligence report is uploaded to an S3 bucket, and a public URL is returned via the initial webhook, providing instant access.

Setup Guide

Prerequisites

Requirement	Type	Purpose
n8n instance	Essential	Workflow execution platform
LlamaIndex API Key	Essential	For robust document parsing and text extraction
OpenAI API Key	Essential	For creating text embeddings and powering the GPT-5-mini AI agent
Pinecone API Key	Essential	For storing and retrieving vector embeddings
AWS S3 Account	Essential	For secure storage of generated PDF reports

Installation Steps

1. Import the JSON file to your n8n instance.
2. Configure credentials:
   • LlamaIndex: Create an "HTTP Header Auth" credential with x-api-key in the header and your LlamaIndex API key as the value.
   • OpenAI: Create an "OpenAI API" credential with your OpenAI API key. Ensure the credential name is "Sumopod" or update the workflow nodes accordingly.
   • Pinecone: Create a "Pinecone API" credential with your Pinecone API key and environment. Ensure the credential name is "w3khmuhtadin" or update the workflow nodes accordingly.
   • AWS S3: Create an "AWS S3" credential with your Access Key ID and Secret Access Key.
3. Update environment-specific values:
   • In the "Upload to S3" node, ensure the bucketName is set to your desired S3 bucket.
   • In the "Create Public URL" node, update the baseUrl variable to match your S3 bucket's public access URL or CDN if applicable (e.g., https://your-s3-bucket-name.s3.amazonaws.com).
4. Customize settings:
   • Review the prompt in the "Analyze" (Langchain Agent) node to adjust the AI's persona or required queries if needed.
5. Test execution:
   • Send sample documents (PDF, DOCX, XLSX) to the webhook URL (/webhook/dd-ai) to verify all connections and processing steps work as expected.

Technical Details

Core Nodes

Node	Purpose	Key Configuration
Webhook	Initiates workflow with document uploads	Path: dd-ai, HTTP Method: POST
Split Multi-File (Code)	Splits binary files, generates unique deal ID	Parses filenames from body or binary, creates dealId from sorted names.
Parse Document via LlamaIndex	Extracts structured text from various document types	URL: https://api.cloud.llamaindex.ai/api/v1/parsing/upload, Authentication: HTTP Header Auth with x-api-key.
Monitor Document Processing	Polls LlamaIndex for parsing status	URL: https://api.cloud.llamaindex.ai/api/v1/parsing/job/{{ $json.id }}, Authentication: HTTP Header Auth.
Insert to Pinecone	Stores vector embeddings in Pinecone	Mode: insert, Pinecone Index: poc, Pinecone Namespace: dealId.
Data Retrieval (Pinecone)	Enables AI agent to search due diligence documents	Mode: retrieve-as-tool, Pinecone Index: poc, Pinecone Namespace: {{ $json.dealId }}, topK: 100.
Analyze (Langchain Agent)	Orchestrates AI analysis using specific queries	Prompt Type: define, detailed role and 6 mandatory Pinecone queries, Model: gpt-5-mini, Output Parser: Parser.
Generate PDF (Puppeteer)	Converts HTML report to a professional PDF	Script Code: await $page.pdf(...) with A4 format, margins, and 60s timeout.
Upload to S3	Stores final PDF reports securely	Bucket Name: poc, File Name: {{ $json.fileName }}, Credentials: AWS S3.
If (Check Namespace Exists)	Implements caching logic	Checks stats.namespaces[dealId].vectorCount > 0 to determine cache hit/miss.

Workflow Logic

The workflow begins by accepting multiple files via a webhook. It intelligently checks if the specific "deal" (identified by a unique ID generated from filenames) has already had its documents processed and embedded in Pinecone. This cache mechanism prevents redundant processing, saving time and API costs. If a cache miss occurs, documents are parsed by LlamaIndex, their content vectorized by OpenAI, and stored in a Pinecone namespace unique to the deal.

For analysis, a Langchain Agent, powered by GPT-5-mini, is instructed with a specific persona and a mandatory sequence of Pinecone queries (e.g., company overview, financials, risks). It uses the Data Retrieval tool to interact with Pinecone, synthesizing information from the stored embeddings. The AI's output is then structured by a dedicated parser, transformed into a human-readable HTML report, and converted into a PDF. Finally, this comprehensive report is uploaded to AWS S3, and a public access URL is provided as a response.

Customization Options

Basic Adjustments:

• AI Prompt Refinement: Modify the Prompt field in the "Analyze" (Langchain Agent) node to adjust the AI's persona, introduce new mandatory queries, or change reporting style.
• Output Schema: Update the JSON schema in the "Parser" (Langchain Output Parser Structured) node to include additional fields or change the structure of the AI's output.

Advanced Enhancements:

• Integration with CRM/Dataroom: Add nodes to automatically fetch documents from or update status in a CRM (e.g., Salesforce, HubSpot) or a virtual data room (e.g., CapLinked, Datasite).
• Conditional Analysis: Implement logic to trigger different analysis paths or generate different report sections based on document content or deal parameters.
• Notification System: Integrate with Slack, Microsoft Teams, or email to send notifications upon report generation or specific risk identification.

Use Case Examples

Scenario 1: Private Equity Firm Evaluating a Target Company

Challenge: A private equity firm receives dozens of due diligence documents (financials, CIM, management presentations) for a potential acquisition, needing a rapid initial assessment.
Solution: The workflow ingests all documents, automatically parses them, and an AI agent synthesizes key company information, financial summaries (revenue history, margins), and identified risks into a structured report within minutes.
Result: The firm's analysts gain an immediate, comprehensive overview, enabling faster screening and more focused deep-dive questions, significantly accelerating the deal cycle.

Scenario 2: M&A Advisor Conducting Preliminary Due Diligence

Challenge: An M&A advisory firm needs to provide clients with a quick, consistent, and standardized preliminary due diligence report across multiple prospects.
Solution: Advisors upload relevant prospect documents to the workflow. The AI-powered system automatically extracts core business model details, investment thesis highlights, and customer concentration analysis, along with key financials.
Result: The firm can generate standardized, high-quality preliminary reports efficiently, ensuring consistency across all client engagements and freeing up senior staff for strategic analysis.

---

Quick Answers

Which documents can it read?
PDF, DOCX, and XLSX via webhook upload, parsed by LlamaIndex.
How does caching work?
A unique deal ID (from filenames) maps to a Pinecone namespace. Repeat uploads skip re-parsing.
Where does the webhook listen?
/webhook/dd-ai. POST sample documents there to test end to end.

Made by: khmuhtadin
Category: AI | Tags: Due Diligence, AI, Automation, M&A, LlamaIndex, Pinecone, GPT-5-mini, Document Processing
Need custom workflows? Contact me

Connect with the creator:
Portfolio • Workflows • LinkedIn • Medium • Threads

Additional info

Built with n8n. Need an assessment on your business? Feel free to reach out at https://khmuhtadin.com/consultation/